107 lines
3.8 KiB
Python
107 lines
3.8 KiB
Python
import os
|
|
import json
|
|
import asyncio
|
|
from dotenv import load_dotenv
|
|
from langchain_openai import ChatOpenAI
|
|
from langchain_core.prompts import PromptTemplate
|
|
from langchain_core.output_parsers import PydanticOutputParser
|
|
from pydantic import BaseModel, Field
|
|
from langchain.tools import tool
|
|
from deepagents import create_deep_agent
|
|
from deepagents.backends import FilesystemBackend
|
|
from langchain_core.messages import HumanMessage
|
|
|
|
# Загрузим переменные окружения (ключ OpenRouter)
|
|
load_dotenv()
|
|
|
|
# 1. Модель карточки задания
|
|
class TaskCard(BaseModel):
|
|
title: str = Field(description="Title of the assignment.")
|
|
subject: str = Field(description="Subject or topic.")
|
|
deadline_hint: str = Field(description="Short hint about the deadline.")
|
|
deliverable_type: str = Field(description="What to submit (report, code, presentation, etc.).")
|
|
grading_hints: list[str] = Field(description="List of hints about grading criteria.")
|
|
|
|
# 2. Парсер
|
|
parser = PydanticOutputParser(pydantic_object=TaskCard)
|
|
|
|
# 3. Шаблон промпта
|
|
prompt = PromptTemplate(
|
|
template="""Extract the following fields from the task description:
|
|
title: the title of the assignment
|
|
subject: the subject or topic
|
|
deadline_hint: a short hint about deadline (e.g., 'by Friday')
|
|
deliverable_type: what to submit (report, code, presentation, etc.)
|
|
grading_hints: list of hints about grading criteria
|
|
|
|
Return the data in JSON format as follows:
|
|
{format_instructions}
|
|
The description: {description}
|
|
""",
|
|
input_variables=["description"],
|
|
partial_variables={"format_instructions": parser.get_format_instructions()},
|
|
)
|
|
|
|
# 4. LLM (OpenRouter)
|
|
llm = ChatOpenAI(
|
|
model="openai/gpt-oss-20b:free",
|
|
base_url="https://openrouter.ai/api/v1",
|
|
api_key=os.getenv("OPENAI_API_KEY"),
|
|
temperature=0.0,
|
|
)
|
|
|
|
# 5. Цепочка: промпт → LLM → парсер
|
|
chain = prompt | llm | parser
|
|
|
|
# 6. Инструмент для парсинга
|
|
@tool
|
|
def parse_task(description: str) -> TaskCard:
|
|
"""Parse a task description into structured fields."""
|
|
return chain.invoke({"description": description})
|
|
|
|
# 7. Бэкенд и агент
|
|
backend = FilesystemBackend()
|
|
agent = create_deep_agent(
|
|
model=llm,
|
|
tools=[parse_task],
|
|
backend=backend,
|
|
system_prompt="You are a task parsing agent. Use the parse_task tool to extract structured data from the assignment description.",
|
|
)
|
|
|
|
# 8. Основная функция
|
|
async def main():
|
|
description = (
|
|
"Сдайте к пятнице мини-отчёт по LangChain: 2 страницы, упор на агентов. "
|
|
"Оценка: за полноту и за пример кода."
|
|
)
|
|
|
|
# Вызов агента
|
|
result = await agent.ainvoke(
|
|
{"messages": [HumanMessage(content=description)]},
|
|
{"configurable": {"thread_id": "session-1"}},
|
|
)
|
|
|
|
# Последнее сообщение содержит вывод инструмента
|
|
output_text = result["messages"][-1].content
|
|
|
|
# Попытка распарсить JSON
|
|
try:
|
|
parsed_dict = json.loads(output_text)
|
|
parsed = TaskCard(**parsed_dict)
|
|
except json.JSONDecodeError:
|
|
# Если вывод не в JSON, используем eval (не рекомендуется в продакшене)
|
|
parsed = eval(output_text)
|
|
|
|
# Вывод валидированного объекта
|
|
print("Валидированный объект:")
|
|
print(parsed.model_dump())
|
|
print("\nКраткая сводка:")
|
|
print(f"Задание: {parsed.title}")
|
|
print(f"Тема: {parsed.subject}")
|
|
print(f"Срок: {parsed.deadline_hint}")
|
|
print(f"Сдача: {parsed.deliverable_type}")
|
|
print(f"Критерии оценки: {', '.join(parsed.grading_hints)}")
|
|
|
|
if __name__ == "__main__":
|
|
asyncio.run(main())
|